Re: [PATCH 0/11] Per-bdi writeback flusher threads #4

[Jens Axboe <jens.axboe@oracle.com>]

On Mon, May 25 2009, Richard Kennedy wrote:
> Hi Jens,
> 
> I've been testing this from your git version which builds as
> 2.6.30-rc6-00057-g81eabcf.
> 
> Unfortunately it's not doing too well.
> 
> When building a kernel with 'make -j 8' on my AMDX2 64bit, the screen
> repeatedly locked up for several minutes at a time,and my music player
> also froze.
> In total the full kernel build took over 80 minutes, normally it's only
> about 15.
> 
> However the machine seems to have recovered correctly, & now everything
> is back to normal.

Weird, perhaps you hit an unlucky revision. I only use the git branch
for development, and it's continually rebased to collect and split
patches and fixes. So I don't generally recommend to use that, just the
posted patches. I build -j8 or larger kernels with the writeback patches
all the time, and haven't seen any issues. That's on a core 2 quad. Just
for kicks, can you send me your .config?

I'll post a new revision tomorrow, if you could try that I'd appreciate
it!

> Maybe it does need the congestion handling after all?

No it does not, by the very nature of the bdi threads being blocking,
congestion is not relevant.

-- 
Jens Axboe